The problem
A data citation can become ambiguous when a provider later revises or withdraws an observation. A chart alone does not tell a reader which release supplied a figure or what changed afterward. CiteGrid gives each complete import a durable snapshot and lets a briefing point to that exact version.
What CiteGrid does
- Compares total, rural, and urban electricity access across Nigeria, Ghana, and Kenya by observation year.
- Publishes a new snapshot only when all nine country and indicator series validate together.
- Shows changed, added, and withdrawn values between adjacent snapshots in a revision ledger.
- Pins the example briefing to a snapshot and exports the cited observations with source information.
- Marks simulated fixture replay and example data visibly so a visitor does not mistake them for a current data feed.
Key decisions
Publish complete evidence
The import validates all nine series before publication and writes a complete snapshot transactionally. A failed import retains an audit record while leaving the previously published snapshot available. This prevents a comparison from mixing values from partial releases.
Keep citations attached to their source
Normalized observation records produce a content hash. A pinned briefing refers to a particular immutable snapshot, while later imports create their own versions and revision records. The reader can inspect both the cited value and what changed after it was pinned.
Make the demonstration boundary explicit
The public site runs on a separate SQLite demonstration database seeded with simulated values. Labels on Explore, About, Revisions, the briefing, and its export identify this boundary. The database is suitable for trying the interface, not for quoting as a live World Bank measurement.
Verification and limitations
The offline test suite found 121 tests: 120 passed and one PostgreSQL-only concurrency test was skipped on SQLite. The public PythonAnywhere deployment was checked with a populated Explore page, revision and provenance pages, fixture replay, and an example briefing. The repository documents the import command, data constraints, and setup steps.
The hosted demo uses SQLite and simulated values. The PostgreSQL concurrency path is implemented but was not verified by the local SQLite run. The free PythonAnywhere site also requires a monthly renewal by its owner, so its availability depends on that upkeep.